Sometimes inspiration can strike from the most unexpected places. It can result in a cross-pollination between ideas commonly used in one field but applied to a completely different one. That might have been the case with a recent paper on lightsail design from researchers at the University of Nottingham that used techniques typically used in video games to develop a new and improved structure of a lightsail.
The core of the paper was the development of an algorithm to design, test, improve, and iterate on the design of a solar sail. Except the focus of this paper wasn’t the traditional reflective solar sail that simply uses light to push itself. These were transmissive sails that refractive the light that hits them, allowing them to generate thrust parallel to its surface rather than perpendicular to it. Transmissive sails are key for applications like orbital station-keeping, as they allow a sail to even raise its orbit without constantly re-orientating itself.
To simulate how the light is refracted by the sail, Samuel Thompson, a recent PhD graduate at the university, turned to a technique common in most modern 3D video games - ray tracing. This computationally intensive approach simulates individual rays of light and determines how they interact with their environment - in this case a solar sail. The key to the technique is simulating lots and lots of these rays all at once, though Dr. Thompson has to remove some rays that fell below a certain energy threshold so that the simulation didn’t overwhelm itself with an ever-increasing number of them.
Fraser discusses the advantages of solar sails with Dr. Slava TuryshevOnce the simulation was working, Dr. Thompson used a second technique commonly found in video games, but on the other end of the spectrum. Reinforcement learning is a machine learning technique commonly used to find the desired result, like beating a video game or increasing the force on a light sail, by trial and error. Plenty of YouTube videos exist that showcase what this looks like in video games, where an algorithm-controlled character makes slightly better progress each “generation” of itself. The same general idea went for the reinforcement learning algorithm to develop the structure of the solar sail.
After the algorithms were complete, Dr. Thompson and his co-authors tried it on two different solar sail structures - a prism and a “lightfoil”. Prisms were extremely efficient - by themselves. But when placed together into a structure that would be used as an actual solar sail, they were disrupted by “pattern propagation” where the refracted light from one prism would fall on another one, causing it to lose some of its propulsive power. To get around this problem, the optimization algorithm used a technique called “pattern skimming” where the prisms were angled so their refracted light was disbursed just past the tip of the adjoining prism. This resulted in a 58% increase in tangential pressure (i.e. thrust), a significant improvement over a previous prism-based design.
Lightfoils aren’t designed for thrust, though. They are designed for stability, which manifests as “corrective torque” that forces the sail to hold its position. This is useful for orbital applications where it is important not to have a spacecraft end up in a catastrophic spin. To solve this problem, the optimization algorithm ended up with a shape that looked like a “rounded pentagonal prism”. Despite its weird design, the resulting structure improved the self-correcting torque by 74% compared to the original semi-cylindrical design of a lightfoil.
An example of a Q-learning algorithm, the technique used to optimize the solar sail in the paper, learning one of the most classic video games - Snake. Credit - CodeBullet YouTube ChannelHowever, the downside was a decrease of 22% in the range of angles where a pattern of those foils remained stable, which is potentially devastating in real-world applications. However, the real advantage of this algorithm was its adaptability. When tasked with finding the best configuration for a real-world deployment on a CubeSat with a different center of mass, it came up with a completely different triangle-like prism shape that increased the torque by 147%.
That adaptability means it will be useful in future designs as well, though as of now the algorithm doesn’t account for other important parts of lightsails, like material selection, weight, and deployment. But the nice thing about optimization algorithms is that they can optimize for whatever parameter is selected - which can be more than one of them. So as solar sail design continues to improve, developing tools to help with that process will become increasingly important.
Learn More:
University of Nottingham - New study shows potential for improved fuel-free spacecraft sails
S.M. Thompson et al. - Modelling and numerical optimisation of refractive surface patterns for transmissive solar sails